high performance computing on graphics processing units: hgpu.org

A tremendous amount of work has been conducted in content-based image retrieval (CBIR) on designing efficient index structure to accelerate the retrieval process. Most of them improve the retrieval efficiency via complex index structures, and few take into account the parallel implementation of algorithm on underlying hardware. It makes the existing index structures suffer from low-degree of parallelism. Therefore, it is of great importance to improve the efficiency of CBIR by designing hardware-adaptive index structure to exploit the advantages of parallel acceleration on hardware. Motivated by this practical need, in this paper, a novel graphics processing unit (GPU) adaptive index structure, termed as plane semantic ball (PSB), is proposed to reduce the work of retrieval process and simultaneously exploit the parallel acceleration of underlying hardware. In PSB, semantics are embedded into the generation of representative pivots and multiple balls are selected to cover more informative reference features. With PSB in place, the online retrieval of CBIR can be factorized into independent components that can be executed on GPU efficiently. Comparative experiments with conventional brute force approach demonstrate that the proposed approach can accelerate the retrieval process with little information loss (1.73x speedup is obtained with 10-4 accuracy loss). Furthermore, PSB is compared with the state-of-the-art approach random ball cover (RBC) on two standard image dataset Corel 10K and GIST 1M, experimental results show that our approach achieves higher speedup than RBC on the same accuracy level.